Track shared memory in metrics (#5023)

* Track shared memory so we can exclude them form resident memory. * Bump psutil version to 3.3.0 to get shared memory metrics.
sigp · Jan 5, 2024 · 0c97762 · 0c97762
1 parent 01994c4
commit 0c97762
Show file tree

Hide file tree

Showing 4 changed files with 18 additions and 32 deletions.
diff --git a/Cargo.lock b/Cargo.lock
diff --git a/common/eth2/Cargo.toml b/common/eth2/Cargo.toml
@@ -36,7 +36,7 @@ pretty_reqwest_error = { workspace = true }
 tokio = { workspace = true }
 
 [target.'cfg(target_os = "linux")'.dependencies]
-psutil = { version = "3.2.2", optional = true }
+psutil = { version = "3.3.0", optional = true }
 procfs = { version = "0.15.1", optional = true }
 
 [features]

diff --git a/common/eth2/src/lighthouse.rs b/common/eth2/src/lighthouse.rs
@@ -243,6 +243,8 @@ pub struct ProcessHealth {
     pub pid_mem_resident_set_size: u64,
     /// The total virtual memory used by this pid.
     pub pid_mem_virtual_memory_size: u64,
+    /// The total shared memory used by this pid.
+    pub pid_mem_shared_memory_size: u64,
     /// Number of cpu seconds consumed by this pid.
     pub pid_process_seconds_total: u64,
 }
@@ -277,6 +279,7 @@ impl ProcessHealth {
             pid_num_threads: stat.num_threads,
             pid_mem_resident_set_size: process_mem.rss(),
             pid_mem_virtual_memory_size: process_mem.vms(),
+            pid_mem_shared_memory_size: process_mem.shared(),
             pid_process_seconds_total: process_times.busy().as_secs()
                 + process_times.children_system().as_secs()
                 + process_times.children_system().as_secs(),

diff --git a/common/warp_utils/src/metrics.rs b/common/warp_utils/src/metrics.rs
@@ -14,6 +14,10 @@ lazy_static::lazy_static! {
         "process_virtual_memory_bytes",
         "Virtual memory used by the current process"
     );
+    pub static ref PROCESS_SHR_MEM: Result<IntGauge> = try_create_int_gauge(
+        "process_shared_memory_bytes",
+        "Shared memory used by the current process"
+    );
     pub static ref PROCESS_SECONDS: Result<IntGauge> = try_create_int_gauge(
         "process_cpu_seconds_total",
         "Total cpu time taken by the current process"
@@ -90,6 +94,7 @@ pub fn scrape_process_health_metrics() {
         set_gauge(&PROCESS_NUM_THREADS, health.pid_num_threads);
         set_gauge(&PROCESS_RES_MEM, health.pid_mem_resident_set_size as i64);
         set_gauge(&PROCESS_VIRT_MEM, health.pid_mem_virtual_memory_size as i64);
+        set_gauge(&PROCESS_SHR_MEM, health.pid_mem_shared_memory_size as i64);
         set_gauge(&PROCESS_SECONDS, health.pid_process_seconds_total as i64);
     }
 }